Last month I opened a file I had built the week before and could not explain why it worked.
It was mine. My project, my repo, my name on the commits. I had shipped it, it passed, it went out.
And sitting there rereading it, I realized I had never actually understood it. I had steered it. The agent wrote it, I nudged, it worked, I moved on.
A week later it was a stranger's code with my name on it.
If you build with AI, you have had this moment. You shipped something fast, it worked, everyone was happy. Then a few days later you open it to change one thing and you are reading it for the first time. Your own system, and you are a tourist in it.
That is the part nobody warned me about.
The speed is real. I am not going to pretend it is not. Things that took me a week take an afternoon now. The demos are faster, the first version is faster, the momentum carries you.
But speed and understanding turned out to be two separate purchases. I thought I was buying both. I was buying one.
Here is the naive version of me from a year ago. I believed that if the output was good, I had done the work. If it ran, I understood it. If it shipped, it was mine.
None of that was true.
Good output is not the same as understanding. Running is not the same as knowing why. And shipping something you cannot explain is not ownership, it is a loan you have not read the terms on.
The bill comes later. It always comes at the worst time.
It comes when production breaks at eleven at night and you are debugging a system you never actually read. It comes when a client asks why it works this way and you hear yourself improvising. It comes when you need to change one thing and you cannot, because you do not have the mental model, you only have the artifact.
I have felt each of those. The eleven at night one more than once.
So here is the opinion, plainly.
The real risk of building with AI is not that it writes bad code. It usually writes fine code. The risk is that it lets you ship code you never understood, and it hides that gap behind a green checkmark. The danger is not failure. The danger is fluent, passing, production-ready work that you cannot actually account for.
Speed makes the gap invisible. Everything is working, so why would you slow down to understand it. And then one day the thing that is working is the thing you have to fix, and understanding is not optional anymore.
I do not have a clean fix, because the honest answer is not a trick. It is a decision.
The decision is to treat understanding as a separate thing you buy on purpose, with time you deliberately spend, even when the code already works. To read the thing the agent wrote before you ship it, not after it breaks. To be able to explain your own system out loud, in your own words, with the file closed.
That is slower. It feels like friction on top of speed you already have. Every part of the current moment is telling you not to bother, because it works, ship it, move on.
I moved on for months. I am paying that back now, one late night at a time.
The thing I keep coming back to is that competence used to be a side effect. You could not ship something without understanding it, because you had to write every line. The understanding came free, welded to the doing.
That weld is gone now. You can ship without understanding. Which means understanding is no longer automatic. It is a thing you have to choose, and most of us have not noticed that the choice exists yet.
I did not notice for a long time. I was too busy being fast.
So the question I would actually put to you, and I mean it as a real question, not a lead-in.
How much of what you shipped this week could you explain, right now, with the file closed?
Not how much works. How much you understand. If those two numbers are the same, you are doing something I have not figured out yet, and I would like to know how.
If they are not the same, you are where I am. Fast, productive, and quietly renting understanding you have not paid for.
The speed is worth it. I would not go back.
But I stopped pretending the two came together. They do not. And the sooner you buy the second one on purpose, the smaller the bill gets.
Your turn
How much of what you shipped this week could you explain with the file closed?
If this was useful
I work through this in public, the wins and the freezes both, mostly on LinkedIn and YouTube. If the real version of building in the open is useful to you, that is where it lives. Find me on X, GitHub, and the work at next8n.com.
Top comments (0)